FedCB: Joint Client Selection and Bandwidth Allocation for Unreliable Aerial Federated Learning

Unmanned aerial vehicles (UAVs) have been widely used to perform search and tracking tasks in military and civil fields. As an emerging distributed learning paradigm, federated learning (FL) is more suitable for UAV networks with computing modules on board. UAVs train local models while a terrestrial base station (BS) aggregates the global model, forming an air-ground collaborative federated learning (AGCFL) system. However, UAVs typically face challenges such as unreliable communication links and limited onboard resources, adversely affecting overall FL performance. To address these issues, we propose a novel FL framework with joint client selection and bandwidth allocation strategy, called FedCB, which aims to minimize the global loss function. Specifically, the strategy mitigates the negative impacts of unreliable communication links by selecting UAVs with superior wireless channel conditions while rationally allocating bandwidth resources. We have designed an alternating iterative optimization-based algorithm to solve this problem. Extensive experimental results demonstrate that compared with other baseline schemes, our solution achieves significant improvements in model accuracy, fully validating its effectiveness and superiority. © 2025 IEEE.